Related to #14651 (upstream issue this fixes) and prior attempt #13190.

Thanks for flagging overlap with #13190.

#13190 was closed as partially redundant: the duplicate-row work landed on main via _section3_emitted_pairs, while the live OpenAI count enrichment path from that PR was not merged. Issue #14651 tracks the remaining symptom (OpenAI / OpenAI Direct still showing total_models: 0 on latest main).

#14753 is intentionally narrower than #13190: it adds a native openai curated catalog in _PROVIDER_MODELS, a small providers: fallback when the base URL is api.openai.com with no explicit models, and an optional /v1/models probe in provider_model_ids("openai") — without re-landing the already-resolved dedupe changes from #13190.

Happy to adjust if maintainers prefer a different approach, but this is meant as the #14651 follow-up, not a re-submission of the full #13190 bundle.

Thanks for this fix, @XieNBi! These changes have already landed on main — this PR's commit was applied directly to the branch.

Automated hermes-sweeper review confirms the implementation is present:

• _PROVIDER_MODELS["openai"] catalog entry added at hermes_cli/models.py line 145 (commit 4a51ab61)
• api.openai.com fallback for empty user-provider model lists at hermes_cli/model_switch.py line 1283
• Optional /v1/models probe in provider_model_ids("openai") at hermes_cli/models.py line 1814
• Regression tests present in tests/hermes_cli/test_user_providers_model_switch.py line 204

The fix is pending inclusion in the next tagged release. Closing as implemented on main.